User Tools

Site Tools


development

Differences

This shows you the differences between two versions of the page.

Link to this comparison view

Both sides previous revision Previous revision
Next revision
Previous revision
development [2018/03/16 20:26]
yzan [Data types]
development [2018/03/25 17:15]
yzan [Data types]
Line 6: Line 6:
   - @baseDate: a single column that is defined as Target Date.    - @baseDate: a single column that is defined as Target Date. 
   - @baseTarget:​ column(s) that is/are defined as Target Column (multiple independent targets are permissible).   - @baseTarget:​ column(s) that is/are defined as Target Column (multiple independent targets are permissible).
 +  - @baseFold: a column with numbers 0..9 for 10-fold cross-validation.
   - @numericalColumn:​ a name of some numerical attribute in @propagatedTable.   - @numericalColumn:​ a name of some numerical attribute in @propagatedTable.
   - @nominalColumn:​ a name of some categorical attribute in @propagatedTable.   - @nominalColumn:​ a name of some categorical attribute in @propagatedTable.
Line 13: Line 14:
   - @targetName:​ a single column from @baseTarget. Intended for Weight of Evidence.   - @targetName:​ a single column from @baseTarget. Intended for Weight of Evidence.
   - @targetValue:​ a value of @targetName column. Intended for Weight of Evidence.   - @targetValue:​ a value of @targetName column. Intended for Weight of Evidence.
 +  - @targetValuePrior:​ a prior probability of the @targetValue. Intended for Weight of Evidence.
  
  
Line 66: Line 68:
   * Boolean type is automatically casted to {0,1} if treated as numerical.   * Boolean type is automatically casted to {0,1} if treated as numerical.
   * Not all functions may work with Longvarchar and longnvarchar types. This is database specific.   * Not all functions may work with Longvarchar and longnvarchar types. This is database specific.
-  * We can group by xml, floating point numbers and timestamps, but the gain from that in well typed dbs is questionable.+  * We can group by xml, floating point numbers and timestamps, but the gain from that in well typed dbs is questionable.  
 +  * We may extract year, month, day in week,... from timestamps and only then treat timestamps as nominal. 
 +  * We may treat timestamps as numerical by calculating datediff(@temporalColumn,​ @baseDate) if @baseDate is available, otherwise calculate datediff(@temporalColumn,​ "​1.1.2000"​).
 ==== Naming convention for patterns ==== ==== Naming convention for patterns ====
   * bare name, if possible, to keep the names short   * bare name, if possible, to keep the names short

Page Tools